Slashdot Mirror
Promised Platform-Independent GPU Tech Is Getting Real
Vigile writes "Last year a small company called Lucid promised us GPU scaling across multiple GPU generations with near-linear performance gains without restrictions of SLI or CrossFire. The company has been silent for some time, but now it is not only ready to demonstrate the 2nd generation hardware, but also to show the first retail product that will be available with HYDRA technology. In this article there is a quick look at the MSI 'Big Bang' motherboard that sports the P55 chipset and HYDRA chip and also shows some demos of AMD HD 4890 and NVIDIA GTX 260 graphics cards working together for game rendering. Truly platform-independent GPU scaling is nearly here and the flexibility it will offer gamers could be impressive."
If it is essentially just a load-balancer, why can't it be done in software?
The article only mentions DirectX, no word about OpenGL, so it must be not a pure hardware solution. If all it does is re-routing of D3D calls, why CPU can't do it?
Lucid is offering up scaling between GPUs of any kind within a brand (only ATI with ATI, NVIDIA with NVIDIA)
Strange... Is the difference between a 10-years-old NVIDIA card and a current-year NVIDIA card really smaller than the difference between a current ATI card and a current nVidia card?
HermesPod: Free Podcast Download Manager for Windows
Let me be the first to saw aw s e! Maybe I can finally get decent full screen flash performance on linux now!
e om
Be relentless!
Finally, we can have asynchronous GPU pairing? And cross-brands to boot? What's incredible is having heard nothing about this for so long, TFA now says a product may hit the market in the next 30 days. I take it that by sidestepping Crossfire and SLI technology, they won't have to pay any licensing fees to either AMD or NVIDIA. Hopefully the patent trolls won't be able to find any fodder that would prevent and delay commercial release.
>>Finally, we can have asynchronous GPU pairing?
I think NVIDIA has some sort of asymmetrical SLI mode available on its mobos with built-in video cards. It allows the weak built in card to help a little bit with the big video card installed in the main PCI-E slot.
IIRC, it gives a 10% boost or so to performance.
Ah, here it is...
http://www.nvidia.com/object/hybrid_sli.html
I know I speak anonymously when I say that this Anonymous Coward does not speak for the rest of us Anonymous Cowards.
So, parent poster, gargle our collective balls.
From the article, "The HYDRA technology, as it is called, is a combination of hardware (in the form of a dedicated chip) and software (a driver that sits between the OS and DirectX)", I can't wait for this software technology to be available for GNU/Linux. But.. something tells me it will take a while as never ATI and NVidia chips can not even do 3D using free software as of today and support seems to be years away. And yes, I know, there is some unstable proprietary binary blob available for my ATI card which can do 3D, but it is immoral to use that and it is actually so slow on 2D (which to me is more important) compared to the free "radeon" driver that it's ridiculous.
9/11: Never forget it was a false-flag operation
There are very good reasons to assume that this story sounds too good to be true. The main claim to fame of the Hydra solution is that separate objects are rendered on separate GPU's and then merged together later. That's really beautiful, but it requires one GPU to access the Z buffer of the other. That's typically not possible or, if it is, only at very low performance. It also doesn't explain how rendered textures will work, which has always been known as one of the main issues for SFR-type multi-GPU rendering (which the Hydra is also.)
Hydra really doesn't intrinsically do anything that couldn't be done by ATI or Nvidia. In fact they have less information at their disposal. E.g. they don't know the post-transform vertex data, which is necessary to know the extent of new objects.
I'm very sceptical that this will be a universal solution that's going to put existing GPU vendors to shame and that's even if those won't put active road blocks in Lucid's way.
Can we ban humans from the internet/electronics by making it so that if they touch something 5 volts of greter they get a VERY hard shock?
With their proprietary CUDA and Firestream technologies, I would think Nvidia and AMD/ATI resepctively would be able to make a daughter card that could add or increase GPU capability on their existing respective hardware, or open up 3rd party licensing to build this market segment.
My ATI X1300 handles far more BOINC than it does games, and I have no real reason to upgrade right now. But if there was an add-on that ATI or an approved 3rd party manufacturer developed that was reasonably priced, I wouldn't hesitate to add functionality.
I seem to remember a similar concept years ago with the Intel 386 SX architecture where you could purchase an optional math co-processor to plug in next to it.
Truly platform-independent GPU scaling is nearly here and the flexibility it will offer gamers could be impressive.
But this is not any ware close enough.
There are going to be some performance hits compared to native crossfile/sli implementations. There are three models of the Hydra 200 part, and they each differ in their pcie lanes. The high-end model, which is going on the MSI motherboard, sports two x16 pcie lanes from the chip to the graphics cards (configurable as 2x16, 1x16 + 2x8, or 4x8), but only a single x16 lane from the chip to the pcie controller. So, where a good high-end crossfire or sli board will have two x16 pcie lanes from the controller to the slots for the gpus, this solution will be limited to one x16, limiting the bandwidth available to each graphics card. Exactly how much of a performance hit this would incur remains to be seen, and it probably depends on the cards being used (an older 8000 series geforce doesn't need/won't use as much bandwidth as a gtx 295, for example), but I would expect as gpus grow more powerful and require more bandwidth to keep them fed and working, we will start to see performance deterioration compared to the native crossfire and sli implementations (although lucid can always modify their design to keep pace).
Incidentally, the two lower-end hydra chips will sport a x8 connection to the controller and 2 x8 connections to the cards, and a x16 connection to the controller and two x16 connections to the cards (strictly 2x16, not configurable in any other arrangement)
The distribution engine as it is called is responsible for reading the information passed from the game or application to DirectX before it gets to the NVIDIA or AMD drivers.
So presumably it will work only in Windows, and only with DirectX games (e.g. not with OpenGL.) I'm guessing that supporting OpenGL would require a big programming effort so we won't see it soon if at all. I suspect there aren't many OpenGL games out there anyway, but I don't follow such things.
Unless the OS market changes drastically, I expect we'll never see non-Windows drivers.
(Aside: ...combine the performance of an AMD and NVIDIA card in the same system. This is truly the killer feature that will make you want, no NEED, a HYDRA-based motherboard very soon. My GPU selection strategy is to buy the most powerful GPU available ... that needs no power connector and is passively cooled. I don't think I'm in the market segment they are addressing here.)
Quattuor res in hoc mundo sanctae sunt: libri, liberi, libertas et liberalitas.
Ah, here it is... http://www.nvidia.com/object/hybrid_sli.html
Im pretty sure that also got discontinued with the 9xxx generation of nVidia GPU's
It'll be interesting to see how much extra latency the chip adds to the rendering process. I don't imagine the hardcore gamers would be too happy about it if they sacrifice an extra 50 ms to gain some FPS.
Articles and articles I've read about Hydra is that it comes in both the HW and the SW parts, with the SW sits in between DirectX and the OS
My point is why only DirectX?
Wouldn't this be cutting itself short --- and be dependent on Microsoft?
While I know that in the world we live now DirectX pwns, but if we don't offer others a chance, how are they gonna be popular enough to rival DirectX ?
Muchas Gracias, Señor Edward Snowden !
Great, now I can have 2 buggy display drivers installed at the same time, each with their own quirks. And who helps me out when I have graphical problems in a game? Do you really think ATI or NVIDIA will give end-user support for this? What about game developer support? It is a support nightmare for all involved. No thanks. Sorry, this idea is brain-dead long before it hits the shelf.
It would make sense if they developed a spec with a common access API to the HW instead of using wrappers like OpenGL/DX on proprietary drivers. HW should expose a platform-independent API so driver could be written by Microsoft or apple or who ever. And by the way, why should I put two different vendors cards on the same machine instead of using the native single vendor solution? It is only useful when OpenGraphics comes out.
eh. SLI/crossfire has always been a niche market. Buying 1 top of the line nvidia or ATI card is always a stronger solution then buying two mid level cards. So this would only make sense if you are buying 2 top of the line cards and honestly while the charts make it look impressive, it's just that, a bragging right. You don't see human detectable improvement in performance in most games/apps. It's a very small market. I believe that's why nidia or ATI hasn't done any real development of their own products to allow a older card to work with a newer card to work together. It seems like a good market idea, increase brand loyalty, if they had a product, knowing you can add a new card a year later and keep the one you already have, the trick being you'd have to buy the same brand again. But no GPU maker has any main stream stuff like this. Why? Because it's fr ricer rigs. GPU's gain performance so fast.
First it was Nvidia, then Nvidia's control over Ageia (of PhysX chip fame)
Now it's MSI's turn in their control over Hydra
What this means is, there are only ATI and Intel out there who are seriously dabbling with graphics hardware, who are not based in Taiwan !
Muchas Gracias, Señor Edward Snowden !
http://www.virtualgl.org/ Will scale across many GPUS on the same board or across the world.
A 10% boost is not really worthing bothering about to be honest. Discrete graphics is so much faster than integrated you might as well turn off the integrated graphics completely.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
Really? I thought we were supposed to be a collective hive mind.
last time i heard of
such magical product
it was april fool
The world belongs to those who get up early. - I'm far from being the king of Earth then
10% isn't a big deal? There are people who go to crazy extremes just to tweak out an extra 1-3% with entire sub markets dedicated to them, so yeah 10% is worth it.
The Goal: A long simple life filled with many complex toys.
From TFA (emphasis mine):
To accompany this ability to intelligently divide up the graphics workload, Lucid is offering up scaling between GPUs of any KIND within a brand (only ATI with ATI, NVIDIA with NVIDIA) and the ability to load balance GPUs based on performance and other criteria.
So what is the deal? Is it cross-brand or not?
Also, they are only planning to launch their chipset on one motherboard with one manufacturer. It all sounds like a short-lived gimmick to me.
A software based solution to the problem of aggregating a heterogeneous collection of parallel OpenGL command streams into one, compositing the output of several graphics cards into one image, or both, has been available for years : It's called Chromium.
Although originally designed for a networked cluster with one gpu per machine, it can conceivably be adapted to one machine with multiple GPUs. Because Chromium's software based compositing would bog down a single processor system, a natural extension would be to build a PCI-E card, running some sort of embedded Unix with a dedicated high speed processor, which would handle compositing the output of multiple GPUs in parallel.
I'm surprised no one has done this yet for Linux.
jdb2
Wow, an actual Haiku! It referenced a season (I'm willing to count "joking season in spring")!
Thanks for your post, it brought a smile to my morning.
Yes? No? Yes? No!?
Dude it's about parallel processing, not quantum computers.
Rampant carbon sequestration destroyed the Dinosaurs' tropical paradise. I'm here to help repair the damage.
Would this technology enable me to use the onboard IGP (Radeon HD3300) which
is now doing absolutely nothing as I am using a separate Radeon HD3850.
Interesting. But let me guess its only compatible with windows.
We anonymously approve and add our numerous collective balls to the group.
J. Doe, president of the Society of Anonymous Cowards Knitters Supermachos (S.a.c.k.s.)